Right now, you have to choose one oversampling of the PSF, you if you have 6x for most of IMCOM then the PSF wing subtraction step also is 6x. However, the far features of the PSF wings don't need the full accuracy, so probably 4x would be fine (though we should test this). It would make IMCOM overall ≈10% faster because we are doing a lot of FFTs at the higher sampling rate.
This requires:
- Putting the appropriate keyword/value in the config file (e.g.,
IMSUBSTRACT_OVERSAMP) with default to the same as the input PSF if not provided.
psfsplit needs to generate the kernel K at the requested sampling and put it in the PSF Legendre polynomial file.imsubtract needs to figure out the appropriate oversampling and read the appropriate kernel from the cached PSF file.
Right now, you have to choose one oversampling of the PSF, you if you have 6x for most of IMCOM then the PSF wing subtraction step also is 6x. However, the far features of the PSF wings don't need the full accuracy, so probably 4x would be fine (though we should test this). It would make IMCOM overall ≈10% faster because we are doing a lot of FFTs at the higher sampling rate.
This requires:
IMSUBSTRACT_OVERSAMP) with default to the same as the input PSF if not provided.psfsplitneeds to generate the kernel K at the requested sampling and put it in the PSF Legendre polynomial file.imsubtractneeds to figure out the appropriate oversampling and read the appropriate kernel from the cached PSF file.